Главная
Коллекция "Otherreferats"
Психология
The influence of outstanding object memories on gist representations in long-term memory
The influence of outstanding object memories on gist representations in long-term memory
The role of gist representations in visual long-term memory. The binary versus the continuum of memory. Consideration of basic units in memory systems. Analysis of hierarchies of visual information. Saliency and summary statistics in visual processes.
| Рубрика | Психология |
| Вид | дипломная работа |
| Язык | английский |
| Дата добавления | 07.12.2019 |
| Размер файла | 280,0 K |
Отправить свою хорошую работу в базу знаний просто. Используйте форму, расположенную ниже
Студенты, аспиранты, молодые ученые, использующие базу знаний в своей учебе и работе, будут вам очень благодарны.
Размещено на http://www.allbest.ru/
Москва 2019
The Influence of Outstanding Object Memories on Gist Representations in Long-Term Memory
Table of contents
memory continuum visual information
Introduction
1. The role of gist representations in visual long-term memory
1.1 The binary versus the continuum of memory
1.2 Basic units in memory systems
1.3 Hierarchies of visual information
1.4 Saliency and summary statistics in visual processes
2. Experimental study of the influence of outstanding objects on gist representations in visual long-term memory
2.1 Methodology
2.2 Procedure
2.3 Results
2.4 Discussion
2.5 Findings
Conclusions
References
Introduction
Memory is one of the fundamental cognitive entities in human psychology. Visual long-term memory - the ability to store and retrieve visual information that is not being currently perceived, and has not been perceived for a long time - is necessary for all psychological and everyday events in the human life, from recognizing our friends and family to being able to find a route in a familiar environment. As the existence of visual memory is quite obvious for anyone who has it (one can simply close their eyes and report the things around them, or the layout of their childhood bedroom, to ensure that they do, in fact, have the ability to store and retrieve information), it has been described much earlier than the beginnings of the cognitive science or, for that matter, psychological science in general.
It has been pointed out that historically research of memory has been focused on defining separate memory systems and measuring their capacity (Brady, Konkle, & Alvarez, 2011). Moreover, memory studies are often based on the usage of the 2AFC (two alternatives forced choice) tasks, in which the participant can give one of the two answers, most often - whether a certain object is old or new. This paradigm has limited sensitivity when it comes to studying not only the fact of correct retrieval, but the characteristics of the information being retrieved, as well as the changes that happen while it is stored.
An alternative research paradigm, in which memory of object features, such as color, can be measured continuously, has been proposed by Wilken and Ma (2004) and later refined by Zhang & Luck (2008) with the usage of mixture models in data analysis. Furthermore, there is now significant evidence pointing towards hierarchical structures of representations as the basis for organization of information in both working and long-term memory of visual knowledge (Brady & Alvarez, 2011; Brady, Schacter, & Alvarez, 2018; Utochkin & Brady, 2018). The bottom level of these hierarchies in visual long-term memory contains feature representations, such as an object's color, size, or orientation, the top level consists of gist representations - that is, representations that contain information regarding statistics of a group of objects, such as the mean color or the standard deviation of size. Object representations take their place in the middle of the hierarchy.
It has been proposed that gist representations are partially responsible for false memories, as well as for adaptive bias in true memories (Brady et al., 2018; Brainerd & Reyna, 2002). Specifically, the memory for individual objects appears to be biased toward the mean of the group of objects. Brady and colleagues propose that bias toward gist is caused by an adaptive mechanism that is aimed at the minimization of error across trials instead of minimization of error for each individual object. However, the process of gist computation in long-term memory remains unclear. On one hand, the base assumption would be that gist is the product of simple averaging of all objects. On the other hand, evidence from ensemble summary statistics in visual perception points toward other possibilities - namely, that outstanding objects will either disproportionately influence the gist, or will be completely excluded from the gist (Haberman & Whitney, 2010). The research regarding memory for salient objects supports the former assumption (Santangelo, 2015).
Thus, the research question at the heart of the present study is the process of gist formation for heterogeneous groups of objects. The described experiment uses the color report paradigm combined with mixture model analysis in order to check the hypothesis that the inclusion of outstanding objects in the memory of a group of objects will disproportionately bias the gist representation toward these outstanding objects. The apparent theoretical discord regarding the influence of outstanding objects on summary statistics in the field of visual processes makes the described study relevant to the field of visual cognition.
The main goal of the presented research is to specify the process of gist computation in the visual long-term memory. It is further divided into theoretical, methodological, and empirical tasks: analyzing the main concepts of basic units and informational structures in long-term memory, as well as the main findings regarding knowledge structures in visual processes, developing the experimental design of an empirical study, replicating the results presented by Brady and colleagues (2018), and examining the hypothesis that outstanding objects will disproportionately influence gist representations in long-term memory. The object of the described research is visual long-term memory. The subject, consequently, is the process of gist computation in visual long-term memory.
The proposed study is novel to the area of visual long-term memory because of the use of the color report paradigm with real-life objects as stimuli in order to deepen the understanding of gist nature. Furthermore, the results obtained in this study have practical applications in education as well as clinical psychology, specifically, in the field of memory impairment diagnostics.
This thesis consists of 1 theoretical chapter, 1 empirical chapter, 35 pages, references not included, and 5 figures, and is based on 30 sources, all of them foreign.
1. The role of gist representations in visual long-term memory
1.1 The binary versus the continuum of memory
Most widely accepted theories of memory, such as Norman's (1969) or Shiffrin & Atkinson's theories (1969), divide memory into three separate subsystems: sensory register, short-term, or working, memory, and long-term memory.
In their paper, Shiffrin & Atkinson propose a memory model that consists of a sensory register, a short-term storage, and a long-term storage. Information is copied from the short-term storage to the long-term storage after it has been in the short-term store for a period of time due to rehearsal, encoding, or decision-making processes regarding that information. According to the model proposed by Shiffrin & Atkinson, long-term memory is akin to a library shelving system based on book contents. In that analogy, books with related contents will be stored in proximity, and so will memories with similar characteristics - for example, similar perceptual qualities, temporal traits, or semantic categories. In this concept of self-addressing memory, both storage and retrieval have in their foundation the same algorithm, a kind of “master plan” specifying where different kinds of information should be stored. Thus, when there is a need to extract data from long-term memory, for example, in a recognition task, memory addresses the specific location where such content should be stored, much the same way the content was, supposedly, stored during memorization. As a result, the saliency of stimuli will influence the speed and accuracy of search in long-term memory: items with highly salient traits will be identified more quickly and easily. According to the theory proposed by Shiffrin & Atkinson, items that are not retrieved from long-term memory are not destroyed: the memories themselves do not decay but are instead not found in the storage. The book is still in the library, safe and sound, but the librarian cannot find it on the shelf.
As pointed out by Brady, Konkle & Alvarez (2011), historically the research of memory has been focused on diving into memory systems and, consequently, measuring each system's capacity. That comes as no surprise, since one of the most famous studies of memory and the creation of one of the pioneer's of memory research, Hermann Ebbinghaus's study of the forgetting curves, donned exactly this approach.
Modern research of memory capacity relies heavily on the 2AFC (two alternatives forced choice) task, in which the participants distinguish between `old' and `new' object. That task contributed to our understanding of memory and other cognitive processes in a myriad of ways. At the same time, seeing as the 2AFC task treats each memory's state as binary (something is either remembered or forgotten, with no in-between), it is not sufficiently sensitive to answer the questions concerning memory organization, the structure of information in the memory storage, as well as the interaction and transformation of memories.
It has long been demonstrated that long-term memory is not simply binary in its nature, but can instead be represented as a continuum from `clearly remembered' to `completely forgotten'. For example, in Binford & Gettys' experiment (1965), the subject chose an answer among alternatives. If on the first try they did not answer correctly, they were given a second chance, and on the second try their answers were correct above chance level.
Wilken and Ma (2004) proposed a color report paradigm that allowed to continuously measure working memory of color. The participants were presented with displays of different color squares. After the presentation, a square cue appeared, centered around the location of one of the previously presented stimuli, indicating which object's color should be reported. A color wheel appeared in the center of the screen, and the participants were instructed to report the color of the object by choosing it on the wheel. Later, Zhang and Luck ( 2008) proposed using a mixture model to parametrize the data collected in the color report task. The mixture model is a combination of two distributions: a uniform distribution meant to represent guesses - the trials in which the participant did not remember anything about the color of the object, and a von Mises distribution, an analog of the normal distribution for a circular space. The von Mises distribution represents the trials in which participants do remember the object's color but the fidelity of their memory is not absolute, which leads to errors. The standard deviation (SD) of that von Mises distribution is used to operationalize memory fidelity - that is, how noisy a representation can be for it to still be retrieved.
In a series of experiments, Brady, Konkle, Gill, Oliva, and Alvarez (2013) applied the color report paradigm to real-world objects and used it to study fidelity in perception, short-term memory, and long-term memory.
The long-term memory condition consisted of two stages. First, the subjects were presented with 232 images of colored real-world objects. Each object belonged to a unique category that did not have a typical color, such as dresses, chairs, or candles. For each trial, the color of the object was generated by rotating each pixel of the image by a random degree in the CIE Lab color space. Each image was presented for either 3 s (experiment 1a) or 1 s (experiment 1b), with a 1 s blank between images. The participants were instructed to memorize each object's identity and color. After the presentation, the participants' color memory was tested in a color report paradigm: all the objects from the study phase were, again, presented in a randomized order. The images appeared in grayscale, with a color wheel surrounding the image. As the participants moved the cursor along the wheel, the color of the image changed, and they were able to adjust the color of the image to their memory by clicking on the location of the color they remembered. The true color of each object, as well as the answer given the by the participant, was represented as the degree between the center of the color wheel, the location of the color on the wheel, and zero, which was located in red. The error between the true color and the answer was calculated and used to fit a mixture model using a maximum likelihood estimation. The results obtained by Brady and colleagues show that, at least for features such as color, representations can be more or less precise, while still being successfully retrieved from memory.
One of the basic assumptions in the foundation of Brady and colleagues' experiments is that objects are not stored in memory as bound indivisible units. In the research of object recognition, it has been presumed that the memory's basic unit is an object: after all, the vast majority of things we perceive and interact with in the real world are, indeed, objects (DiCarlo & Cox, 2007). However, recent studies of memory structure present compelling evidence of the contrary.
1.2 Basic units in memory systems
In working memory research, one of the most frequently debated issues is the question regarding basic units of memory. Luck and Vogel (1997) presented evidence that visual working memory stores bound objects, rather than separate features. The authors conducted a series of experiments, in which they studied the participants' working memory for features, objects, and feature-feature conjunctions (in the form of objects with features from different dimensions, such as color and orientation). The subjects were presented with a sample array and, after a blank, a test array of simple shapes with varying features, and then answered whether the test array was different from the sample array. The first set of experiments used colored squares of the same size as stimuli. The accuracy was operationalized as a function of the set size. The accuracy was high for sets of up to 3 objects, and declined significantly when set size increased from 4 to 12 squares. Similar results were obtained for orientation memory. Adding the verbal load (two digits presented before the sample array which the participants were required to remember and announce after the test array), as well as varying the duration of the sample array presentation and changing the experimental paradigm in order to limit decision-making factors, did not significantly influence the results. Furthermore, additional dimensions in the task, such as a requirement to remember both the orientation and the color of all the stimuli in the array, did not change the 4 object limit of memory accuracy. Thus, the authors make a conclusion that the basic unit of working memory is an integrated object: eight objects with eight features (for example, eight shapes of eight different colors) will be remembered less accurately than four objects with eight features (for example, four rectangles of four different colors and orientations).
However, consequent research (Olson & Jiang, 2002) did not support what is called a `strong-object hypothesis': that is, the hypothesis that the capacity of visual short-term memory is only limited by the number of objects to be held in memory. The authors demonstrated that short-term memory capacity for color is primarily limited for the number of colors to be remembered, and not by the number of objects, and, when controlled for feature heterogeneity, memory capacity for single-feature objects is better than that for conjoined-features objects. Olson and Jiang conclude that visual short-term memory limit depends on both the number of objects and the number of features, thus supporting a `weak-object hypothesis'. As Fougnie and colleagues point out (Fougnie, Cormiea, & Alvarez, 2012), it is possible that additional objects require additional resources to store them, which is another explanation for the results obtained by Luck and Vogel (1997) besides a strong-object hypothesis. Furthermore, if the object-benefit is the result of an integrated object memory units, then feature memories should have the same dynamics: forgetting one feature should mean that the other features of the same object are forgotten as well. The experiments conducted by Fougnie and colleagues show that this is not the case. Different features of the same object, such as shape and color of simple stimuli, are forgotten independently.
Brady and colleagues (Brady, Konkle, Alvarez, & Oliva, 2012) further studied forgetting rates of different features of the same items using images of real-life objects, such as couches, chairs, and cars. They applied the same logic to the long-term memory: if object features are stored as separate, independent representations, the forgetting rate of separate features will be independent as well.
In the first experiment, Brady and colleagues presented the participants with real-world objects of different colors and states, such as closed and open boxes or folded and unfolded chaise lounges, and tested color and state memory after a short or a long delay. The results of this experiment show different forgetting rates for color and state memories. Thus, the color was forgotten significantly faster than the state in real-life objects. In the following experiment, the authors looked at the dependence between state and exemplar: that is, how likely it is that the participant will remember the exact box (exemplar) if they remember correctly that the box was open or closed (state). Memory for both state and exemplar for tested after a short delay (30 minutes) and after a long delay (3 days). The results from these experiments demonstrate that memories for different object features in long-term memory are forgotten independently - which supports the idea that objects are not a basic indivisible unit of information in memory. Additional evidence supporting that claim has been presented by Utochkin and Brady (2018).
1.3 Hierarchies of visual information
In place of the described object-based model organization of visual information, a hierarchical model has been proposed (Brady et al., 2011). According to that model, both working and long-term memory consist of multiple levels of representations, with feature representations at the bottom of the hierarchy, object representations in the middle, and gist representations at the top. Gist representations here are defined as an entity that contains summary statistics about a group of objects, united because of their similarities.
Gist representations in the hierarchy are similar to the concept of `images', or `codes', in the theory of Shiffrin & Atkinson (1969): that is, groups of information that are related to each other and likely to be retrieved together. They specify in their 1969 paper that `images' should not be comprehended as a single unit of information: what may form an 'image' in one task would not do so in a different process. Atkinson and Shiffrin base the existence of `codes' on their concept of location-addressable, or, more specifically, self-addressable long-term memory.
That definition is similar to the concept of gist representations, as used by Brady, Schacter, & Alvarez (2018): that is, a representation of a group of objects united, for example, because of their categorical classification. It should be noted that, according to Brady and colleagues, such gist representations are not a bound group of object representations remembered together, but rather a representation that contains descriptive statistics for that group, such as its mean size, color, or orientation. In a way, Atkinson and Shiffrin's 'images' may be considered a predecessor of the gist representations concept in long-term memory.
A possible mechanism for the creation of such `gist representations' is memory consolidation. Cognitive and neurophysiological studies, led by the Mьller-Pelzer memory preservation and consolidation hypothesis (McGaugh, 2012) demonstrate that memory storage is an active, not passive, process, during which memories transform and consolidate.
Fuzzy-trace theory (Brainerd & Reyna, 2002) includes a similar structure of verbatim (individual) and gist (group) memories and proposes that gist memories serve as a basis for false memories. This claim has been supported by Brady, Schacter, and Alvarez (2018).
In their first experiment, the participants were presented with 5 blocks of 40 objects from 4 categories (10 objects from each category), presented in random order. Each category had a randomly chosen color `gist': for example, all lamps could be blue-green, and all backpacks pink-purple. The categories were chosen so that they did not have a `typical' color.
After each block, the participants were tested on their object and color memory: 20 objects from the presentation, as well as 20 new objects from the same categories, were once again presented in random order. They appeared on the screen in grayscale, and the participants had to answer whether each object was old or new, and then, regardless of the answer, they had to adjust each object's color in a color report task. In the second experiment, the same stimuli were used, but this time they were presented in 10 study block, with each block comprised of 10 objects from the same category. The test trials were the same as in the first experiment, with two changes: firstly, the participants were not required to report the color if they answered that the object was new. Secondly, after reporting the color, the participants were required to assess their confidence in their memory of the object's color, from 1 (`totally guessing') to 6 (`extremely sure').
The analysis of data from both experiments showed that the participants' color memory was biased toward the mean color of the object's category. Additionally, the bias was lower when the participants' confidence in their memory was high. However, even when the subjects were very confident in their memory (5 or 6 on the scale of confidence), their color report was still biased toward the categorical mean. Similar results have been obtained for working memory (Brady & Alvarez, 2011). The authors explain it as an adaptive process based on a Bayesian integration model: this distortion of individual memories toward categorical gist, while resulting in a small error in each individual case, leads to a minimized average error.
The exact process of gist computation in long-term memory is yet unclear. It is reasonable to presume that gist is computed linearly, with each object representation equally contributing to the gist. A similar process, during which summary statistics of a group of objects is computed, has been studied in visual perception and visual working memory (Brady & Alvarez, 2011; Utochkin & Brady, 2019). The resulting computation is called ensemble statistics. The influence of outstanding stimuli has been a topic of research in connection to the ensemble statistics.
1.4 Saliency and summary statistics in visual processes
In the field of visual perception (Ariely, 2001; Chong & Treisman, 2003), it has been shown that, when presented with a set of objects, such as a display of differently sized circles, the subjects cannot differentiate between members and non-members of the set but are quite successful in the mean-discrimination task, in which they are required to answer whether a sample circle is bigger or smaller than the mean of the set. Furthermore, Chong & Treisman's (2005) experiments indicate that such means are computed automatically and in parallel, and can be successfully computed for groups segregated by color and location.
Haberman & Whitney (2009) showed that ensemble coding occurs even for high-level perceptual features, specifically, emotions: the participants of their experiments successfully discriminated the mean emotion in the sets of 16 faces even when the presentation time was as low as 500 ms, which shows that ensemble coding is fast and possible at multiple levels of visual perception, and not for the lower-level features only. Additionally, the subjects were significantly less accurate when looking at distorted faces, which supports the claim that such ensemble processing is not caused by the analysis of low-level perceptual properties of the faces. In a consequent experiment, Haberman and Whitney (2010) demonstrated that the visual system excludes emotionally outstanding stimuli when computing the emotional mean of a set of faces.
Recent limited capacity models, on the other hand, (Kanaya, Hayashi, & Whitney, 2018) propose that the most salient items in the ensemble are averaged preferably. Furthermore, the research of saliency in memory shows that salient, both perceptually and semantically, objects are remembered better compared to less salient objects (Santangelo, 2015). This is in accord with Atkinson and Shiffrin's theory (Shiffrin & Atkinson, 1969), as well as the research conducted by Konkle and colleagues (Konkle, Brady, Alvarez, & Oliva, 2010).
In the experiments conducted by Konkle and colleagues, the participants were presented with 2800 images of real-life objects divided into categories, with different number of objects contained in each category:1, 2, 4, 8, or 16 objects. After the presentation, the subjects' object memory was tested in a 2AFC-task. The results of the experiments show that categorical distinctiveness influences the probability of retrieval in long-term memory: as the number of objects in a category increased, the hit rate decreased.
Seeing as there is evidence supporting hierarchical encoding in the long-term memory, a question regarding the computation of gist representation arises. It is possible that salient objects will disproportionately influence gist representations or, on the contrary, will be excluded from gist computation altogether. Evidence from the research of ensemble statistics in visual perception shows that both outcomes are possible. Previous research on gist representations indicates their adaptive role in memory and supports the idea of memory as a reconstructive process, in which individual memories are influenced by each other. At the same time, some of the most valuable information regarding cognitive processes is the information about its mistakes (Schacter, 1999). As Schacter argues, mistakes of memory are not actually mistakes of the process, they are a byproduct of that process, and give us valuable information regarding that process. In much the same way, possible `mistakes' of gist computation are a promising source of information regarding the intricate details of memory processes. The described evidence from the research of ensemble statistics in visual perception provides two possible `mistakes' in the gist computation: either the gist will be computed with the exclusion of outstanding objects, or it will be preferentially biased toward these objects. Memory research supports the second claim, seeing as salient objects are remembered better. A more thorough understanding of gist computation in long-term memory would provide valuable insight into the organization of long-term memory.
2. Experimental study of the influence of outstanding objects on gist representations in visual long-term memory
2.1 Methodology
Recent research regarding the structure of knowledge in visual processes supports the idea of a hierarchy of representations (Brady & Alvarez, 2011; Chong & Treisman, 2005; Utochkin & Brady, 2019), with low-level sensory representations at the bottom, object representations in the middle, and representations of summary statistics of object groups at the top. Such summary statistics in visual long-term memory are called `gist representations' because they contain a `gist' of information from a group of objects joined by the similarity of their characteristics, such as their categorical classification. An influential theory of false memories, the fuzzy-trace theory, proposes that such gist representations are the foundation for false memories (Brainerd & Reyna, 2002), and recent research supports that claim (Brady et al., 2018). On one hand, gist representations' role in false memories, as well as bias toward gist in individual true memories, supports the idea of a Bayesian integration model as the basis for long-term memory adaptation. On the other hand, the existence of gist representations details our understanding of informational structures in long-term memory. However, the specifics of the gist computation are yet unclear. The basic assumption would be that gist is calculated linearly, with each object equally contributing to the summary statistic, such as the mean. At the same time, the existing studies of similar processes in visual perception provide possible support for two other models: a gist in which outliers, or outstanding objects, are not included (Haberman & Whitney, 2010), or a gist that is itself biased toward more salient objects (Kanaya et al., 2018).
The research of salient object memory is in accord with the idea of gist biased toward outstanding objects (Santangelo, 2015). However, the lack of research regarding gist computation specifically in visual long-term memory and the contradicting evidence from other visual processes result in a scientifically relevant research question regarding the nature of gist representations in visual long-term memory.
The experiment proposed in this thesis is novel because it is the first research concerning the relationship between outstanding object representations and gist representations in long-term memory, as well as the first experiment focused on the process of gist computation in long-term memory. Furthermore, the usage of real-world objects in visual research lends ecological validity to the experiment. The practical significance of the proposed research is due to the importance of understanding memory structure and the role of gist in false memories in both education and clinical psychology, specifically, the field of neurorehabilitation.
The main theoretical hypothesis of the presented study is the hypothesis that the inclusion of outstanding objects in the memory of a group of objects will result in a gist representation that is biased toward these outstanding objects.
The object of the described research is visual long-term memory. The subject, in its turn, is the process of gist computation in visual long-term memory.
The main goal of this research is to characterize the process of gist computation of object groups with outstanding objects. It can be further divided into separate objectives:
1) Theoretical objectives
a) Analyze the main concepts of basic units and informational structures in long-term memory.
b) Analyze the main findings regarding knowledge structures in visual processes. Highlight the main studies describing summary statistics representations in visual knowledge.
2) Methodological objectives
a) Develop the experimental design of an empirical study.
3) Empirical objectives
a) Replicate the results of Brady and colleagues (2018)
b) Examine the hypothesis that outstanding objects will disproportionately influence gist representations in long-term memory.
An experimental approach that allows studying memory as a continuum has been proposed by Wilken and Ma (2004). In their proposed paradigm, the participants reported the color of a stimulus - in their case, a square - by clicking the color's location on a color wheel. Each color on the wheel can be computed as the value in degrees of the angle between zero, located in red, the color, and the center of the wheel. Thus, the error for each stimulus can be calculated as the difference between the stimulus' correct color and the answer given by the participants (Fig. 1).
Figure 1 Operationalization of true (a) and chosen (b) colors of stimuli. Error calculated based on true and chosen colors of stimuli (c)
Zhang & Luck (2008) presented a model for the parametrization of the data collected in such a color report paradigm: the vector of errors made by each participant can be used to fit a mixture model, that is, a model of a mix of two distributions (Fig. 2). A uniform distribution represents all the trials in which the participant randomly guessed, and the area of that distribution represents the percentage of guessing. A von Mises distribution, which is a Gaussian-like distribution for a circular space, represents the trials in which the participant did not guess, but instead gave an answer with a certain degree of precision. The standard deviation of the von Mises distribution, thus, represents the fidelity of feature representations in the participant's memory. Brady and colleagues (2013) applied that approach to images of real-life objects, such as couches, buttons, or candles.
Figure 2 Example of a mixture model used to operationalize guessing and fidelity
It has been demonstrated that, when objects belong to categories, and these categories have a typical color throughout the experiment, individual representations are biased toward the typical color of the category (Brady et al., 2018). That can be demonstrated when fitting a mixture model with bias: by default, the von Mises distribution of the mixture model is fit so that the mean is located in zero. However, it is also possible to fit a model in which the mean of the von Mises distribution is not preassigned, but is instead chosen by maximum likelihood estimation, quite the same way as it is done with the area of the uniform distribution and the standard deviation of the von Mises distribution. If the errors are represented in such a way that positive values represent errors toward the categorical color, and negative values represent errors that are biased away from the categorical color, then the positive bias in the mixture model will demonstrate that individual representations are biased toward the mean color of the category. In the same way, bias toward outstanding objects can be computed by fitting a mixture model to a vector of errors in which error is calculated so that it is positive when it is biased toward the color of the outstanding objects.
This logic is applied in the described experiment. In this thesis, object representations in long-term memory are understood as hierarchically organized structures of feature representations, with color as the specific feature, which memory is measured in the presented study. Gist representations are understood as representations that contain summary statistics of a group of objects, unified on the basis of their similarity, in this case - the fact that all objects of a group belong to one semantic category.
Bias of the gist toward mean color is operationalized as a bias of the von Mises distribution in the mixture model fit to a vector of errors calculated relative to the mean color of the category, whereas bias toward outstanding objects is operationalized as the same bias of the von Mises distribution, but fit to a vector of errors relative to the color of the outstanding object.
The main hypothesis of the presented study is that gist representations are disproportionately biased toward outstanding objects. The operationalized hypotheses, hence, are as follows:
1. Bias of the von Mises distribution toward the color of outstanding objects will be higher than bias toward the mean color of the category.
2. Bias toward mean will be higher in ordinary categories than in the categories that contain outstanding objects.
3. Bias toward mean color will be higher in ordinary objects than in outstanding objects.
2.2 Procedure
Sample. The sample consisted of 26 participants aged from 18 to 31 (Median age = 22), 22 of them female. All the participants gave their informed consent to participate in the study and had normal eyesight and color vision, as assessed by Snellen charts and Ishihara's test for color deficiencies (1936). The subjects were given a small cash compensation of 150 rubles, which was not conditional on the results of the experiment.
Apparatus. The experiment was run in Octave 4.0 using Psychophysics Toolbox (Brainard, 1997) and the code provided by T. Brady (Brady et al., 2013). The stimuli were presented on CRT monitors with a size of 49°x17°, which were color calibrated with Spyder5ELITE before the conduction of the experiment.
Stimuli. 300 images of real-world objects were used as stimuli. The same images were previously used in the research of Brady and colleagues (2018). The stimuli belonged to 20 different categories, such as lamps, chairs, or baby strollers, with 15 objects in each category. All the categories were chosen so that they did not have a specific color associated with them.
Figure 3 Experimental design of the described study
Stages. The experiment consisted of 5 identical blocks (Fig. 3). Each block had two stages: the presentation stage and the testing stage. During the presentation stage, the participants were shown 40 categorical objects for 4 categories, 10 objects from each category. Each object appeared on the screen for 2 s. Color of the objects from each category was generated randomly from a normal distribution with a randomly chosen mean and a standard deviation of 30° in the CIE L*a*b colorspace. As a result, all the objects from the same category were colored similarly: for example, the lamps could be green or blue, but never red. In half of the categories across all blocks, one object was an outlier: its color was not a part of the categorical normal distribution, but was located near its tails. For example, for the green-blue lamps, an outlier lamp would be shown in purple or yellow.
During the testing, the participants were presented with the images from the presentation phase along with new objects from the same categories, in random order. Altogether there were 40 old objects from 4 categories (10 objects from each category) and 20 new objects from the same 4 categories (5 objects from each category). During the testing phase, the objects were presented in grayscale. For each object, the participants had to answer whether that object is old or new, by pressing either right or left Shift key on the keyboard. After that, no matter which answer was given by the participant, they were required to remember the color of the object by choosing it on a color wheel. During the instruction, the experimenter explained that sometimes memory makes mistake, and it is possible to think that an old object is new, but be able to retrieve its color. The color wheel was rotated on each trial for a random degree in order to eliminate location-related color bias.
2.3 Results
All the objects presented in the experiment can be divided into two types: objects with a color sampled from a category distribution, which will be called ordinary objects, and the outstanding objects, that is, objects with a color sampled from outside of the category distribution. The mean hit rate for ordinary objects was equal 0.57, and was significantly different from 0.5 (chance), t(25) = 2.50, p =.020. The mean false alarm rate for ordinary objects amounted to 0.34, and the mean sensitivity measure in the signal detection theory (Stanislaw & Todorov, 1999), d', was 0.60.
The probability of hits for outstanding objects was 0.59. It was also significantly different from chance, t(25) = 2.25, p =.033, There was, nowever, no significant difference between hit rates for ordinary and outstanding objects according to Student's t-test for paired samples: t(25) = -0.64, p =.53, d = 1.57.
Only the data from hit trials was used in the subsequent analysis: hits, that is, correctly identified old objects, contain individual object information as well as gist information, and, opposed to false alarms, were both presented in the first stage of the experiment and tested in the second, and, consequently, errors can be calculated for hits, but not for false alarms. Thus, in order to analyze change of gist representation, these types of answers were used.
Errors - the difference between the color in which the objects were presented and the color reported by the participant - were calculated in two different ways, relative to the mean color of the category and relative to the color of the outstanding object. Error calculated relative to the mean color of the category means that the error was calculated in such a way that positive errors are directed toward the mean color of the category (Fig. 4a), whereas negative errors are directed away from the mean color of the category (Fig. 4b). For this type of errors, positive bias in a mixture model shows a bias toward the mean color.
Figure 4 Example of error calculation relative to the mean color of the category: error toward the mean (a) and error away from the mean (b)
In the same way, errors relative to the outstanding object's color were computed so that positive errors were directed toward the outstanding object's color, and negative errors - the other way around. For this type of errors, positive bias in a mixture models shows a bias toward the outstanding object's colors.
Bias from the mixture models for different types of errors and different categories were used in all comparisons.
Figure 5 Means and 95% confidence intervals of bias for different mixture models
In the first comparison, bias toward mean (Mdn = 18.5) and bias toward outstanding objects (Mdn = -0.63) in categories that contain outstanding objects were analysed using Wilcoxon Signed-ranks test, and the difference proved to be significant: Z = 4.10, p <.001, r =.80.
The comparison between bias toward mean in ordinary categories (Mdn = 19.4) and categories containing outstanding objects (Mdn = 18.5) was not significant according to the results of Wilcoxon Signed-ranks test, Z = 0.67, p =.5, r =.13.
The bias toward mean for ordinary objects was compared with the bias toward mean for outstanding objects. In all the other comparisons, data were analysed by subjects. Here, since the number of outstanding objects presented to each participant was 10, which is not enough data for a correct fit of a mixture model, individual subject bias toward mean for ordinary objects were compared with one value - the bias of the model, fit on all outstanding objects across the participants, which was -36.7. Student's t-test for a hypothesized mean showed a significant difference: t(25) = 19.7, p <.001, d = 3.87.
2.4 Discussion
Firstly, it should be pointed out that the memory for outstanding objects was not better than the memory for ordinary objects. The results show that the hit rate - the possibility of correctly identifying an `old' object as old - did not differ significantly for the two type of objects. Thus, the concept of self-addressing memory, as proposed by Shiffrin and Atkinson (Shiffrin & Atkinson, 1969), was not supported in this experiment.
Secondly, the ordinary objects of categories that contained outstanding objects were biased toward the mean color of the category much more noticeably than toward the color of outstanding objects. According to the obtained results, there is no reason to argue that bias toward outstanding objects was present all. The null hypothesis that the mean of the distribution of bias in this case is not equal to zero could not be rejected.
Furthermore, color memory for ordinary objects from different categories did not differ significantly: the null hypothesis that the bias toward mean in categories with and without outstanding objects comes from one distribution could not be rejected. Hence, the hypothesis that the addition of outstanding objects will influence the gist of the category was not supported by the results of the experiment.
The bias relative to the mean color of the category in the memory for outstanding objects themselves, nevertheless, was significantly different than for ordinary objects. More specifically, whereas ordinary object representations were biased toward the category mean, the opposite effect was present for outstanding objects. Their bias was observably directed away from the category mean.
The results of the present study replicate the findings of Brady and colleagues (2018) that memory of individual objects is biased toward the gist. However, the data does not supports the idea that outstanding objects will disproportionately bias gist representations in visual long-term memory, since bias toward mean in objects from ordinary categories and outstanding object categories did not differ significantly.
In fact, quite the opposite effect was present: according to the results, outstanding objects were not included into the averaging process that takes place when gist representations are computed. This finding supports the results obtained in the research of ensemble statistics in visual perception by Haberman and Whitney (2010). The fact that the bias for outstanding object representation was highly negative, that is, outstanding object representations were biased away from the mean color of the category, as opposed to toward it, proposes that outstanding object representations are repulsed away from the categorical mean - that is, they are encoded as more different from the category than they really are. This proposes that, even if outstanding objects are not remembered better than ordinary objects, they are remembered and processed differently and excluded from the gist representation. A similar effect of repulsion was described in the research of object representations interaction in visual working memory (Bae & Luck, 2017). However, in that example, objects with similar features were repelled away from each other, whereas in the discussed results objects with dissimilar features are “contrasted out”.
In their paper, Brady and colleagues (Brady et al., 2018), suppose that the bias toward gist in individual object representations is an adaptive mechanism based on a Bayesian integration model that is aimed at minimizing average error in visual long-term memory. The results obtained in the present study support that explanation and point toward a complex adaptation process at play in visual long-term memory: outstanding objects are not included in the gist and their individual representations are repulsed. Such a strategy is beneficial in the task of differentiating between objects when reporting their features. It ensures that the error is minimized across trials: firstly, the gist is the result of averaging only `representative' objects, which means that the adaptation is beneficial even for groups with outliers. Secondly, the outliers themselves seem to be encoded differently from ordinary representative members of the category and reported with a separate inverted bias. This mechanism does not allow errors caused by a wrong bias. Additionally, this adaptation strategy presumes that false memories - which have been proposed and shown to be based in gist representations (Brady et al., 2018; Brainerd & Reyna, 2002) - are founded in the average object features, with is in agreement with the recent work proposing that false memories themselves lead to benefits in behaviour and problem-solving (Howe, 2011).
Overall, the results of the presented experiment point toward an adaptive nature of visual long-term memory, proposing that the main of this adaptation is not precision of memory for each particular object, but instead a minimization of overall error.
2.5 Findings
1) Replication of the finding that individual object representations are biased toward gist representations in visual long-term memory.
2) Outstanding objects are excluded from gist representations in visual long-term memory.
3) Outstanding objects are repulsed when they are encoded in visual long-term memory.
4) An advanced adaptive strategy is used in visual long-term memory when encoding outstanding objects and computing the gist of a heterogeneous group of objects.
Conclusions
It is commonly assumed that objects are the base unit of information in visual cognition (DiCarlo & Cox, 2007). Notwithstanding, recent research shows that object features are forgotten independently in visual memory (Brady et al., 2012), which, together with evidence regarding binding errors and bias toward summary statistics in individual object memories (Brady et al., 2018; Chalfonte & Johnson, 1996; Utochkin & Brady, 2018, 2019), points to a hierarchy of visual information. In the lower level of it are feature representations, gist representations - that is, summary statistics of a group of objects joined because of a common denominator - are in the top, and object representations are placed in the middle of the hierarchy.
The color report paradigm, proposed by Wilken and Ma (2004) and later modified by Zhang and Luck (2008), allowed to study object memory continuously. It was applied to real-life objects and used to study the interaction between different levels of such hierarchies. The experiments conducted by Brady and colleagues (2018) showed that representations of individual objects are biased toward the gist representation of the group, which the authors explained as an adaptive mechanism based in a Bayesian integration model. The specifics of gist formation in visual long-term memory, however, are in need of further research. Evidence from visual perception, combined with data regarding the role of saliency in memory, proposed that outstanding objects will disproportionately bias gist representations.
Подобные документы
Theoretical basis of a role plays as a teaching aid. Historic background of game origin. Psychological value of a role plays. The main function and principles of game organization. Gaming technique. Classification of role plays. Advantages of a game.
курсовая работа [50,7 K], добавлен 26.04.2013The term charisma has two senses: compelling attractiveness or charm that can inspire devotion in others and a divinely conferred power or talent. Fundamental Secrets of uniqueness and success. How to use the full force of his charisma to succeed.
презентация [1,3 M], добавлен 11.03.2015The definition of stress as the body's way of responding to any kind of stimuli. Consideration of positive and negative emotions, which may cause stress. External and internal causes of stress. The role of consciousness in the assessment of events.
презентация [1,1 M], добавлен 22.09.2015Influence psychology of cognitive activity and cognitive development on student’s learning abilities during study. Cognitive development theory in psychology. Analysis of Jean Piaget's theory. Her place among the other concept of personal development.
презентация [1,3 M], добавлен 13.04.2016Research of negative influence of computer games with the elements of violence and aggression on psychical development of children and teenagers. Reasons of choice of computer games young people in place of walk and intercourse in the real society.
доклад [15,3 K], добавлен 10.06.2014The study of harm to children from watching American cartoons. Problem of imitating negative or mindless characters from cartoons. Leading role of American cartoon industry in the animation history. First steps in the progress of a child’s development.
эссе [16,3 K], добавлен 11.04.2013Тілесні методи боротьби зі стресом. Виявлення стратегій поведінки комбатантів у психотравмувальних ситуаціях як одного з ресурсів у процесі їх подальшої психологічної реадаптації та реабілітації. Теоретичне обґрунтування застосування методики "BASIC Ph".
статья [214,6 K], добавлен 05.10.2017Definition of Leadership. Trait theory. How this theory works. Origin and Analysis and basics Pre-conditions for effective use of Trait theory. Inborn leadership characteristics. Process of impact and interaction among the leader and his followers.
реферат [436,9 K], добавлен 24.09.2014Memory, Teaching and his types. Why we need teaching of memory. Short-term and protracted memory: oppositions and coincidences. Short-term memory and methods of his improvement. Listening of methods is in translation. Scholars of research of listening.
дипломная работа [45,4 K], добавлен 10.07.2009The attitude to veterans. Education of moral and Patriotic feelings in children of preschool age. Let's keep the memory, for veterans, for the future generation. Attitude of my generation to the veterans and the fact that they have done for us.
презентация [9,8 M], добавлен 19.09.2013
